Size feature-serving capacity against peak online lookup demand. Enter peak requests, serving instances, benchmark throughput per instance, and a target utilization to calculate fleet capacity, utilization, headroom, and required instance count.
Use sustained benchmark throughput rather than a short burst maximum.
Use the main result together with the supporting bounds, counts, or capacity figures. The estimate is only as reliable as the input definitions, sampling process, and operating assumptions.
Benchmark with realistic feature payloads, cache behavior, tail latency, network limits, and redundancy. Maximum throughput alone is not a latency guarantee.
14 units at 3,600 requests per second provide 50,400 requests per second. Demand of 42,000 uses 83.3% of raw capacity.
Total capacity equals the number of workers or instances multiplied by measured throughput per worker or instance.
Operating below theoretical maximum leaves headroom for bursts, retries, uneven load, maintenance, and performance variation.
Use a credible peak demand for sizing. Average traffic can materially understate the resources needed during busy periods.
No. It estimates throughput capacity. Validate p95 and p99 latency separately under realistic load.
Not always. Shared databases, queues, networks, coordination, and cache behavior can reduce scaling efficiency.
| Input | Unit |
|---|---|
| Peak demand | Requests or events per second |
| Unit throughput | Sustained requests or events per second |
| Target utilization | Planned operating ceiling |